Split-out PR A from #3584. Normalizes the Antigravity/agy Gemini 3.5 Flash tier IDs to clean public names (gemini-3.5-flash-low/medium/high), maps them to the live upstream IDs at the executor boundary, and removes Antigravity from the global model resolver so the executor owns wire normalization. Maintainer follow-up: kept gemini-3.5-flash-preview as a hidden backward-compat alias routing to the High tier (so saved combos/configs keep working). Live-validated the tier set via the agy CLI catalog. Integrated into release/v3.8.21. Thanks @dhaern!
* chore(release): open v3.8.20 development cycle
* fix(images): prefer bare combos over image aliases (#3527)
Integrated into release/v3.8.20
* fix(translator): map Codex local_shell tool (#3534)
Integrated into release/v3.8.20
* fix(usage): make opencode-go quota fetcher fail-open instead of throwing 500 (#3522)
Integrated into release/v3.8.20
* Fix Runtime page breaker state rendering (#3533)
Integrated into release/v3.8.20
* Expose provider breaker degradation threshold setting (#3535)
Integrated into release/v3.8.20
* fix(executor): strip provider prefix from versioned built-in tool model field (#3532)
Integrated into release/v3.8.20
* feat(providers): add Claude Fable 5 support (#3524)
Integrated into release/v3.8.20
* feat(resilience): add global provider cooldown tracking to prevent combo re-walking (#3556)
Integrated into release/v3.8.20 (default OFF, opt-in)
* fix(translator): scope thoughtSignature bypass to Antigravity/CLI only (#3560)
Integrated into release/v3.8.20. Co-authored-by: Six7Day <six7day@gmail.com>
* fix(routing): normalize thinking:disabled for combo-substituted models that reject it (#3554) (#3563)
Integrated into release/v3.8.20
* fix(usage): accept 0/empty budget limits so the dashboard can save and clear (#3537) (#3564)
Integrated into release/v3.8.20
* docs(changelog): credit @Six7Day for #3560 thoughtSignature fix (#3414)
The #3560 squash co-author trailer landed inline (unparsed by GitHub), so add
an explicit CHANGELOG credit ensuring @Six7Day (original #3414) and @oyi77 are
on the public record for the Gemini thoughtSignature fix.
* fix(gamification): dedup badge unlock via user_badges so events don't re-fire every request (#3472) (#3565)
Integrated into release/v3.8.20
* fix(routing): pass through 'auto' keyword on codex /v1/responses instead of rewriting to codex/auto (#3509) (#3566)
Integrated into release/v3.8.20
* fix(cli-tools): normalize apiKey null in guide-settings schema so cloud-mode config saves (#3552) (#3567)
Integrated into release/v3.8.20
* fix(catalog): reclassify PublicAI from keyless to one-time-initial (requires API key) (#3558) (#3568)
Integrated into release/v3.8.20
* fix(gemini-web): surface missing Playwright browser as actionable 503 + cooldown hint, not a retryable 500 loop (#3516) (#3570)
Integrated into release/v3.8.20
* fix(security): sanitize raw err.message in web executors + embeddings/search response bodies (Rule #12) (#3494, #3495) (#3573)
Integrated into release/v3.8.20
* fix(dashboard): point CustomHostsManager + FeatureFlagsGrid at real routes (#3486, #3487) (#3574)
Integrated into release/v3.8.20
* chore(providers): remove dead krutrim entry (#3483) + docs(api): fix agent-bridge per-agent state route (#3489) (#3575)
Integrated into release/v3.8.20
* docs(api): correct API_REFERENCE.md paths for skills/plugins/admin/cache/acp/system-info (#3497) (#3577)
Integrated into release/v3.8.20
* fix(proxy): drive SOCKS5 UI option from runtime ENABLE_SOCKS5_PROXY, not build-time NEXT_PUBLIC (#3508) (#3579)
Integrated into release/v3.8.20
* fix(playground): filter playground models by node prefix so custom-endpoint models appear (#3505) (#3581)
Integrated into release/v3.8.20
* fix(usage): show an informative message instead of a blank Kiro quota card when no usage breakdown (#3506) (#3582)
Integrated into release/v3.8.20
* docs(changelog): add the #3506 Kiro quota entry (missed in #3582 due to a stale-base CHANGELOG anchor) (#3583)
Integrated into release/v3.8.20
* fix(auto-update): use stable PROJECT_ROOT walker, not frozen process.cwd() (#3561)
Integrated into release/v3.8.20. Auto-update PROJECT_ROOT now uses a stable __dirname-anchored upward walker instead of the no-op process.cwd() resolver.
* fix: address PR #3518 review comments (lifecycle hooks, regex, indentation, route params) (#3562)
Integrated into release/v3.8.20. Addresses #3518 review: regex literals, logs/[id] route params (Next 16), indentation, and wires plugin lifecycle hooks (onInstall/onActivate/onDeactivate/onUninstall) in the loader so manager.ts can register them. Adds Rule #18 regression test.
* docs(changelog): credit @ViFigueiredo (#3423) for PROJECT_ROOT + log #3561/#3562 (v3.8.20)
* fix: openai to gemini incorrectly translates historical tool calls into text (#3569)
Integrated into release/v3.8.20. Standard Gemini direct path now maps historical tool calls to native functionCall/functionResponse parts (signaturelessToolCallMode: native) instead of inert text — validated against the real Gemini API (gemini-2.5-flash returns 200 for signatureless native functionCall, even with tools+thinking; Hard Rule #18). Eliminates the text-serialization leak. Antigravity/CLI sentinel path (#3560) untouched.
* docs(changelog)+test: reconcile standard-Gemini native mode (#3569) — update round-2 rationale comment + log VPS validation
* docs(changelog): reconcile v3.8.20 — add 9 missing bullets + move [Unreleased] to versioned section
* docs(changelog): complete v3.8.20 reconciliation — 27 bullets, 11 contributors
---------
Co-authored-by: Alexander Averyanov <alex@averyan.ru>
Co-authored-by: Hakan Kurşun <bykamaka@gmail.com>
Co-authored-by: Wilson <pedbookmed@gmail.com>
Co-authored-by: Randi <55005611+rdself@users.noreply.github.com>
Co-authored-by: Giorgos Giakoumettis <giorgos@yiakoumettis.gr>
Co-authored-by: PizzaV <103120356+pizzav-xyz@users.noreply.github.com>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: Markus Hartung <mail@hartmark.se>
* chore(release): open v3.8.18 development cycle
* fix(catalog): stop Codex CLI model-catalog refresh from erroring (#3481)
Codex's model-catalog refresh (codex_models_manager) does
GET /v1/models?client_version=<v> and decodes a JSON object with a
TOP-LEVEL `models` array. OmniRoute answers in the OpenAI-standard
`{object,data}` shape, so codex fails with "missing field `models`"
and logs "failed to refresh available models" on every startup.
Detect codex clients via the `originator` / `user-agent` = `codex_*`
headers they send and add an EMPTY top-level `models: []` so the decode
succeeds. Non-codex OpenAI clients keep the byte-identical `{object,data}`
response.
The array is intentionally empty: codex replaces its built-in per-model
agent prompt (`base_instructions`, ~21k chars) with whatever a populated
entry carries for the selected model, so emitting our catalog would drop
the agent prompt to nothing and break codex's agent behaviour (verified
empirically against codex 0.137). An empty list keeps codex on its
built-in model info — same inference as before, minus the error.
Validated end-to-end with the real handler against codex 0.137:
"failed to refresh available models" → 0 occurrences, instructions
preserved (built-in Codex agent prompt, not empty).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: ignore quality reports and local prompt artifacts
Add generated quality gate reports, metrics files, and local setup prompt
artifacts to .gitignore to prevent committing environment-specific or
temporary files.
* fix(provider): detect Responses API format when body has `input` but … (#3490)
Integrated into release/v3.8.18
* fix(sse): normalize numeric provider ids to strings (#3451)
Integrated into release/v3.8.18
* feat(browserPool): resolve Playwright proxy from proxy_registry DB (#3492)
Integrated into release/v3.8.18
* fix(theoldllm): generate X-Request-Token server-side, drop Playwright (#3491)
Integrated into release/v3.8.18
* feat(plugins): add lifecycle hooks and theme-manager plugin (#3473)
Integrated into release/v3.8.18
* fix(combo): parallel pre-screen + circuit-breaker fast-exit for priority combos (#3169)
Integrated into release/v3.8.18
* feat(ui): unifi active and finished requests into single view #1422 (#3401)
Integrated into release/v3.8.18
* docs(changelog): record #3401, #3473, #3492, #3490, #3451, #3491, #3169 under v3.8.18
* feat(docs): add doc accuracy gate + refresh AGENTS.md counts (#3510)
Integrated into release/v3.8.18
* fix(sse): drop empty-choices chunks without usage instead of injecting retry text (#3513)
PR #3422 ('allow OpenAI usage-only empty choices chunks') reintroduced the
assistant-content injection '[OmniRoute] Upstream returned an empty response.
Please retry.' for empty `choices: []` chunks that carry no valid usage. Clients
(Goose/opencode) feed that text back as a turn and spin in a retry loop -- the
exact regression #3400 had fixed by dropping the chunk.
Restore the drop behavior for the no-usage case while preserving #3422's
standards-compliant forwarding of usage-only `include_usage` final chunks.
Realign the mislabeled stream-utils test (it asserted the injection) and add a
dedicated regression guard.
Reported-by: @mochizzan
Refs: #3502, #3388, #3400, #3422
* fix(authz): fall back to URL token when Authorization isn't a usable Bearer (#3504)
Integrated into release/v3.8.18
* fix(playground): authenticate via session, test key policy by id (#3503)
Integrated into release/v3.8.18
* docs(changelog): record #3510, #3504, #3503 under v3.8.18
* fix: llama base url normalization (#3519)
* docs(changelog): reconcile v3.8.18 — add #3519, #3513, #3435-repair, gitignore chore (full commit↔changelog coverage)
* fix(opencode-plugin): bound regex quantifiers in normaliseFreeLabel (polynomial-ReDoS)
CodeQL js/polynomial-redos: unbounded \s* before an anchored \s*$ allowed
O(n²) backtracking on attacker-influenced display names. Bounded to {0,8}/{1,8}
(ample for any real label spacing). Plugin builds + 254 tests green.
* fix(types): restore clean typecheck:core for v3.8.18 release gate
- getPendingRequests() typed to real shape (was widened to object) → fixes
unknown 'count' in the unified-requests view (#3401)
- streamChunks log payload cast to its declared type (callLogs.ts)
- preScreenTargets aligned to canonical IsModelAvailable signature (#3169),
Promise.resolve-normalized so .catch never hits a bare boolean
All 5 gates green: lint(0 err) + typecheck:core + cycles + docs-all + unit + vitest(146).
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Andrey Borodulin <borodulin@gmail.com>
Co-authored-by: Dmitrii Safronov <zimniy@cyberbrain.cc>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: PizzaV <103120356+pizzav-xyz@users.noreply.github.com>
Co-authored-by: Markus Hartung <mail@hartmark.se>
Co-authored-by: Felipe Almeman <4226997+zhiru@users.noreply.github.com>
Claude Code -> claude-opus-4-8 turns intermittently died with 'tool call could not be parsed (retry also failed)'. OmniRoute's claude identity cloak rebuilt the anthropic-beta header from scratch and unconditionally forced interleaved-thinking-2025-05-14 (+ advanced-tool-use / effort for heavy agents), even when the client never negotiated them. The forced interleaved-thinking conflicts with tool_choice-forced turns, producing malformed opus tool_use streams (and sibling 400 'Thinking may not be enabled when tool_choice forces tool use').
selectBetaFlags now takes the client's inbound anthropic-beta: when present, thinking/effort betas are only emitted if the client requested them. Opaque clients (no header — the OAuth cloak path) keep the full set unchanged, so existing behavior and the #2454 model-tier gating are preserved.
Co-authored-by: Forcerecon <Forcerecon@users.noreply.github.com>
* chore(release): open v3.8.15 development cycle
Version bump 3.8.14 -> 3.8.15 (root + electron + open-sse + openapi + lockfiles)
and seed the v3.8.15 changelog placeholder (root + 41 i18n mirrors).
* fix(catalog): add getTokenLimit fallback for combo targets with unknown context (#3369)
Integrated into release/v3.8.15. Fixes applied on the contributor's branch: removed duplicate JSDoc opening in accountFallback.ts and dropped a test asserting unreachable catalog behavior (models with no registry/spec/synced source are filtered before the getTokenLimit fallback at catalog.ts:499).
* fix(combo): add 429 to PROVIDER_FAILURE_ERROR_CODES to prevent infinite retry loop (#3366)
Integrated into release/v3.8.15. Comment block reconciled on the contributor's branch to remove the contradictory 'intentionally excluded' text that remained from the original code.
* fix(auto-combo): include no-auth providers declaratively (#3365)
Integrated into release/v3.8.15. Cleanup applied on contributor's branch: removed duplicate migration 095 (already exists from PR #3338), reverted CHANGELOG.md and i18n changelogs to release versions (release process owns these), dropped package version-bump noise from stale fork base. Core feature — declarative no-auth via serviceKinds metadata, declarative VEO as 'video' provider, anonymousFallback flag for opencode-zen/opencode-go — integrated cleanly.
* fix(migrations): restore 095_provider_node_custom_headers migration
The squash merge of PR #3365 accidentally deleted this migration because
the cleanup commit on the contributor's branch included 'git rm' for the
file (which was a duplicate on their branch). The migration was merged
in v3.8.14 via PR #3338 and must be present in the release branch.
Restoring from git history.
* fix: update Command Code base URL from /alpha/ to /provider/v1/ (#3372)
Integrated into release/v3.8.15.
* feat(error-rules): provider-specific error classification with scope (#3370)
Integrated into release/v3.8.15. PR has genuine value beyond #3369: (1) getProviderErrorRuleMatch now accepts native Headers objects from fetch(); (2) checkFallbackError also uses the provider rule registry — the real end-to-end wiring in the combo fallback path; (3) S4 end-to-end test proving the wiring fires. Merge commit on contributor branch resolved the add/add conflict by taking the #3370 version throughout.
* fix(auto-combo): validate web-session credentials (#3371)
Integrated into release/v3.8.15. Core feature: provider-aware web-session credential validation — hasUsableWebSessionCredential() replaces the broad Object.keys check in virtualFactory.ts, ensuring only sessions with the required storageKeys are included in auto-combo. Cleanup: removed duplicate 095 migration, reverted CHANGELOG/i18n, dropped package bump noise.
* fix(migrations): restore 095_provider_node_custom_headers (deleted again by #3371 squash)
Same issue as after #3365: git rm in the contributor cleanup commit
was included in the squash, deleting this migration from release.
Permanent fix needed: use 'git checkout origin/release -- <file>'
instead of 'git rm' when cleaning up duplicate files in contributor branches.
* fix(kiro): probe Windows %APPDATA%\kiro\storage.db in auto-import (#3363) (#3375)
Integrated into release/v3.8.15. Test fix applied: kiro-windows-auto-import-3363.test.ts now sets DATA_DIR to a fresh temp dir before importing app modules, ensuring isAuthRequired() sees an empty settings DB (no password → auth not required). This fixed test 4 (synthetic SQLite) which was getting 401 due to settings DB state leakage.
* chore(release): finalize v3.8.15 changelog — 2026-06-07
---------
Co-authored-by: Hernan Javier Ardila Sanchez <hjasgr@gmail.com>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: Muhammad Nabil Muyassar Rahman <65392758+TapZe@users.noreply.github.com>
Co-authored-by: kiro-agent[bot] <245459735+kiro-agent[bot]@users.noreply.github.com>
* chore(release): open v3.8.14 development cycle
Version bump 3.8.13 -> 3.8.14 (root + electron + open-sse + openapi + lockfiles).
Seed the v3.8.14 changelog with the four post-tag hotfixes that shipped to
Docker/Electron in v3.8.13 but missed the immutable npm 3.8.13 (#3336 SSRF /
CodeQL #323, #3334/#3335/#3339 Electron packaging). i18n CHANGELOG mirrors get
the in-progress placeholder section.
* feat: add per-provider custom headers support for OpenAI/Anthropic-compatible nodes (#3338)
Integrated into release/v3.8.14
* fix: Kiro Builder ID token import fails with Bad credentials (#3333)
Integrated into release/v3.8.14 — adds Builder ID cached-creds + OIDC refresh path for Kiro token import, with regression tests (#3333).
* Improve code quality: auto-pr/docstrings-1780792063 (#3337)
Integrated into release/v3.8.14 — docstring for context analytics route re-export.
* fix(catalog): remove minimaxai/minimax-m3 from NVIDIA NIM tier (404 upstream) (#3329) (#3341)
NVIDIA NIM does not host minimaxai/minimax-m3 — every request returns
404 page not found, while sibling minimaxai/minimax-m2.7 on the same provider
works. Advertising a model that 404s is a catalog bug; remove it from the nvidia
tier (it remains on the tiers that actually serve MiniMax M3). Re-add only once
NVIDIA serves it.
Co-authored-by: mikmaneggahommie <mikmaneggahommie@users.noreply.github.com>
* fix(cli): write OpenCode config to ~/.config on all platforms incl. Windows (#3330) (#3343)
resolveOpencodeConfigDir used %APPDATA% on Windows, but OpenCode reads its
config from XDG ~/.config/opencode/ on every platform (on Windows:
%USERPROFILE%\.config\opencode\, NOT %APPDATA%). So a Windows user who
configured OpenCode via the dashboard had the file written where OpenCode never
looks — it silently had no effect.
Use the XDG path (XDG_CONFIG_HOME || ~/.config) unconditionally. Update the UI
note + route JSDoc, and flip the three tests that encoded the old %APPDATA%
behavior (t40 per-platform + card-note, cli-runtime-extended getCliConfigPaths).
Co-authored-by: abdulkadirozyurt <abdulkadirozyurt@users.noreply.github.com>
* fix(proxy): make auto-selection fallback opt-in (#3332) (#3344)
selectWorkingProxyFallback (Step 11 of resolveProxyForConnection) listed ALL
registry proxies, ignoring assignments and per-connection proxy_enabled, and
returned the first working one with level:'autoSelect'. So a single proxy added
to the registry silently became a global fallback for every connection's traffic.
Gate it behind a new PROXY_AUTO_SELECT_ENABLED feature flag (default off): the
fallback now no-ops unless the operator opts in. No registry proxy becomes a
silent global default anymore.
Co-authored-by: hertznsk <hertznsk@users.noreply.github.com>
* fix(sse): treat MiniMax M3 as multimodal so vision isn't stripped (#3328) (#3342)
MiniMax M3 via the opencode provider (oc/minimax-m3-free) appeared blind:
image inputs didn't reach the model, while the same model in Cline could
see them. Verified empirically that MiniMax M3 on the opencode upstream IS
multimodal -- a base64 image is described correctly (it returns 403 only
for remote image URLs, which it doesn't accept).
Root cause: OmniRoute treated MiniMax M3 as a non-vision model in two
places, so when compression was active the image was replaced with a text
placeholder before dispatch:
- compression's modelSupportsVision() heuristic (lite.ts) only matched
gpt-4/4o/claude-3/gemini/vision -- minimax was absent -> replaceImageUrls
stripped the image.
- the opencode minimax-m3-free catalog entry lacked supportsVision, so the
combo vision-capability gate could also exclude/mishandle it.
Add 'minimax-m3' to the vision heuristic and supportsVision: true to the
opencode minimax-m3-free entry. TDD: a failing-then-passing test in
compression/lite.test.ts proves replaceImageUrls now keeps images for
minimax-m3 ids, plus a registry assertion mirroring the #2822 qwen test.
Reported-by: @mikmaneggahommie
* docs(i18n): translate 25 core documentation files to Indonesian (#3348)
Integrated into release/v3.8.14 — Indonesian i18n docs.
* fix(review): resolve /review-reviews battery findings (LEDGER-1..11) on v3.8.14 (#3350)
Integrated into release/v3.8.14 — /review-reviews battery hardening (LEDGER-1..11) for #3338 custom-headers + #3333 kiro, plus cycle-test drift fixes (#3329/#3330/#3332).
* fix(provider-proxy): honor per-account proxy toggles (#3349)
Integrated into release/v3.8.14 — honor per-account proxy toggles + auto-fallback opt-in via PROXY_AUTO_SELECT_ENABLED.
* fix(dashboard): remove duplicate Distribute Proxies button on provider page (#3352)
* fix(providers): reduce proxy label noise (#3346)
Integrated into release/v3.8.14 — reduce proxy label noise + a11y (aria-label/sr-only).
* fix(duckduckgo): restore bare Response contract and rebase onto release/v3.8.14 (#3323)
Integrated into release/v3.8.14 — browser-backed cookie providers (duckduckgo/claude-web) with restored executor contract + unit tests.
* fix(noauth): expose only usable model aliases (#3345)
Integrated into release/v3.8.14 — noauth usable-alias filtering + registry alias plumbing (veo-free).
* fix(dashboard): stop infinite config-load loop on Hermes Agent detail page (#3353)
* fix(electron): tree-kill the server on exit/update to release the omniroute.exe lock (#3347) (#3354)
* chore(release): finalize v3.8.14 changelog + clear release-gate drift
- CHANGELOG: finalize the v3.8.14 section (date, full New Features/Bug Fixes/
Maintenance coverage of all 16 cycle commits, Contributors hall of 12).
- docs: document OMNIROUTE_BROWSER_POOL + WEB_COOKIE_USE_BROWSER (#3323) in
.env.example + ENVIRONMENT.md; regenerate the id/llm.txt strict mirror (#3348
had translated it; llm.txt mirrors must match root).
- test(proxy-fetch): #3323 made tlsClient.available a computed getter — stub it
via Object.defineProperty instead of assignment (5 tests were red on the base).
* fix(translator): coerce Gemini functionDeclaration parameters to an OBJECT schema (#3357) (#3360)
* fix(gemini): resolve truncation/suppression of false positive textual tool call markers in backticks (#3358)
Integrated into release/v3.8.14 — Gemini/Antigravity textual tool-call marker normalization (no false-positive suppression + split-chunk buffering).
* docs(changelog): add #3358 Gemini textual tool-call normalization to v3.8.14
* fix(dashboard): surface real analytics error instead of generic placeholder (#3356) (#3361)
The Analytics page discarded the server's error body on a non-OK response and
rendered a generic "An error occurred", so users (and maintainers) could not see
why /api/usage/analytics 500'd after an upgrade. Now the route returns the real
reason via buildErrorBody (sanitized, Hard Rule #12) and the page surfaces it via
a new readFetchErrorMessage helper that handles both the OpenAI-style and legacy
error shapes.
Reported-by: @superti4r
---------
Co-authored-by: PizzaV <103120356+pizzav-xyz@users.noreply.github.com>
Co-authored-by: Someres <168349709+quanturbo@users.noreply.github.com>
Co-authored-by: Dong Mengzhe <154944819+Lang-Qiu@users.noreply.github.com>
Co-authored-by: mikmaneggahommie <mikmaneggahommie@users.noreply.github.com>
Co-authored-by: abdulkadirozyurt <abdulkadirozyurt@users.noreply.github.com>
Co-authored-by: hertznsk <hertznsk@users.noreply.github.com>
Co-authored-by: Krisna Santosa <54174372+KrisnaSantosa15@users.noreply.github.com>
Co-authored-by: Randi <55005611+rdself@users.noreply.github.com>
Co-authored-by: Wilson <pedbookmed@gmail.com>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: Ardem2025 <ardemb22@gmail.com>
* chore(release): open v3.8.13 development cycle
Bump 3.8.12 → 3.8.13 across package.json, lockfile, electron/, open-sse/, and
docs/reference/openapi.yaml; add the [3.8.13] cycle placeholder to the root
CHANGELOG and the 41 i18n mirrors. Integration branch for the v3.8.13 cycle —
fixes/features land here via per-issue PRs and it merges to main at release time.
* fix(ci): skip auto-deploy when VPS host is unreachable from the runner (#3299)
Integrated into release/v3.8.13
* fix(dev): auto-rebuild better-sqlite3 on Node ABI mismatch at dev startup (#3301)
Integrated into release/v3.8.13
* feat(api): accept path-scoped API keys on client API routes (#3300)
Integrated into release/v3.8.13
* fix(sse): harden against empty responses causing Copilot Chat failures (#3297)
Integrated into release/v3.8.13
* fix(api): remove Completions.me rickroll provider (discussion #3293) (#3302)
Integrated into release/v3.8.13
* fix(opencode-provider): extract contextLength from live model catalog (#3298)
Integrated into release/v3.8.13
* feat(web-cookie): self-service login infrastructure + auto-refresh daemon (#3292)
Integrated into release/v3.8.13
* docs(changelog): record the v3.8.13 PRs merged this round (#3292/#3300/#3297/#3298/#3301/#3302/#3299)
* fix(auth): harden URL token extraction — drop query-string fallback, gate to client routes (security follow-up to #3300) (#3309)
Security follow-up to #3300 — integrated into release/v3.8.13
* docs: rename resolve-issues → review-issues skill references
* fix(dashboard): keep no-auth providers visible under 'Show configured only' (#3290) (#3312)
no-auth providers (opencode, duckduckgo-web, theoldllm, veoaifree-web) never
create a DB connection row so stats.total stays 0, which the configured-only
filter treated as 'unconfigured' and hid them — even though they are always
usable and appear unconditionally in /v1/models. filterConfiguredProviderEntries
now treats displayAuthType === 'no-auth' as configured.
Co-authored-by: uniQta <uniQta@users.noreply.github.com>
* fix(cli): resolve update paths relative to script + recursive backup (#3295) (#3313)
omniroute update always failed on a global install:
- getCurrentVersion() read package.json from process.cwd(), which on a global
npm/brew install is the user's working dir, not the package root → null →
'Could not determine current version'.
- createBackup() resolved bin/ from cwd too, and passed the 'cli' directory to
copyFileSync → EISDIR, swallowed by the catch → 'Failed to create backup'.
Both now resolve package.json/bin relative to the script via import.meta.url,
and the backup uses cpSync({recursive:true}) so the cli/ directory is copied.
Co-authored-by: uniQta <uniQta@users.noreply.github.com>
* fix(theoldllm): read upstream body once to avoid [502] body-already-read (#3296) (#3314)
On the cached-token path the executor never enters the refresh branch, so the
same upstream Response was read with .text() twice (token-rejection check +
final body). A Response body is single-use, so the second read threw
'Body is unusable: Body has already been read', caught and surfaced as [502].
Read the body once into finalBody and only re-read after a token-rejection
refetch.
Co-authored-by: onizukashonan14-png <onizukashonan14-png@users.noreply.github.com>
* fix(sse): strip leaked internal tool envelopes from streaming output (#3311)
Integrated into release/v3.8.13
* fix(sse): expose Claude + Gemini budget tiers in the antigravity catalog (#3184) (#3303)
Integrated into release/v3.8.13 (#3184)
* fix(catalog): compute combo context_length from known targets only (#3304)
Integrated into release/v3.8.13 — live contextLength + known-targets combo context (#3298 follow-up)
* chore(i18n): add message keys for proxy UI + vscode/ollama endpoint (#3307)
Integrated into release/v3.8.13 — i18n message keys for proxy UI + vscode/ollama
* feat(dashboard): i18n the proxy settings UI (#3310)
Integrated into release/v3.8.13 — i18n the proxy settings UI
* feat(api): model catalog enrichment + MCP model-catalog tools (#3306)
Integrated into release/v3.8.13 — model catalog enrichment + MCP model-catalog tools, reconciled with #3309 URL-token hardening
* test(catalog): align Antigravity preview-alias test with #3303 budget tiers
#3303 added the Gemini `-high`/`-low` budget tiers to ANTIGRAVITY_PUBLIC_MODELS
(user-callable on the Antigravity OAuth backend, verified via #3184), but did
not update the catalog-route test that asserted `antigravity/gemini-3.1-pro-high`
must NOT be exposed. The assertion now reflects the intended behavior — the
client-visible budget alias IS surfaced — while keeping the legacy
`gemini-claude-*` alias keys unexposed. Caught running the full catalog suite
on the merged release HEAD (the #3303 round only ran the antigravity-aliases
and usage-hardening files).
* docs(changelog): record the 6 PRs merged this review round into v3.8.13
#3306/#3307/#3310 (New Features — VS Code split: catalog+MCP, i18n keys, proxy
UI i18n), #3311/#3303/#3304 (Bug Fixes — SSE envelope sanitizer, antigravity
budget tiers, combo known-targets context_length).
* chore(release): finalize v3.8.13 changelog and cleanup
Finalize the v3.8.13 changelog with release date, maintenance notes,
and contributor credits. Update MCP docs to reference the correct tool
inventory diagram, exclude nested .claude worktrees from ESLint scans,
and tighten a response sanitizer type guard.
* fix(dashboard): refresh connections after provider auth import (#3320)
Integrated into release/v3.8.13 — refresh connections after provider auth import
* fix(codex): strip client-only params on native /responses passthrough (#3317) (#3325)
A /v1/responses request against the built-in codex/ provider does an
openai-responses -> openai-responses passthrough (CodexExecutor.transformRequest
returns the body early for _nativeCodexPassthrough). It forwarded client-only
fields verbatim and the Codex upstream rejected them with 400 Unsupported
parameter: prompt_cache_retention / safety_identifier / user — breaking Factory
Droid (which injects all three). The chat-completions path already strips these
(base.ts #1884, openai-responses translator #2770) but the passthrough skips
translation. Strip the three fields in the shared block before the passthrough
return; user is removed unconditionally since Codex /responses always rejects it.
Co-authored-by: tycronk20 <tycronk20@users.noreply.github.com>
* fix(dashboard): normalize agent-bridge /state response to stop page crash (#3318) (#3326)
The Agent Bridge page seeded a well-shaped initialData default then replaced it
wholesale with the raw /api/tools/agent-bridge/state response. The route returns
{ server, agents } but the UI reads { serverState, agentStates, bypassPatterns,
mappings }, so serverState became undefined and AgentBridgeServerCard crashed on
serverState.running — surfaced as the full-page 'Internal Server Error' boundary
(client render error, not a real 5xx).
Add a shared normalizeAgentBridgeState() that maps the route shape into the page
contract (server.running/certExists -> serverState) and always returns safe
defaults (never undefined serverState). Wired into both the SSR loader (page.tsx)
and the polling hook. The legacy 'agents' entry shape differs from AgentStateEntry
so it is not coerced; full route<->page contract reconciliation (port, upstreamCa,
bypassPatterns, mappings, agentStates) is a follow-up.
Co-authored-by: tycronk20 <tycronk20@users.noreply.github.com>
* docs: VS Code/Ollama endpoints + env & i18n tooling (#3319)
Integrated into release/v3.8.13 — VS Code/Ollama docs + env & i18n tooling
* feat(provider): test-all endpoint, rate-limit overrides, visibility f… (#3267)
Integrated into release/v3.8.13 — provider test-all endpoint, rate-limit overrides, model visibility
* feat: auto-combo optimization, playground model dropdown, only-configured toggle (#3322)
Integrated into release/v3.8.13 — auto-combo candidate expansion + playground dropdown + only-configured toggle
* feat(api): VS Code Copilot Ollama-compatible BYOK endpoint (#3316)
Integrated into release/v3.8.13 — VS Code Copilot Ollama-compatible BYOK endpoint (reconciled with #3306/#3309 auth hardening)
* chore(release): document #3320 in the v3.8.13 changelog + contributor credits
---------
Co-authored-by: Felipe Almeman <4226997+zhiru@users.noreply.github.com>
Co-authored-by: Wilson <pedbookmed@gmail.com>
Co-authored-by: Hernan Javier Ardila Sanchez <hjasgr@gmail.com>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: uniQta <uniQta@users.noreply.github.com>
Co-authored-by: onizukashonan14-png <onizukashonan14-png@users.noreply.github.com>
Co-authored-by: tycronk20 <tycronk20@users.noreply.github.com>
Co-authored-by: Vinayrnani <vinayrnani@gmail.com>
- chipotle/grokTls: explicit null checks instead of a Promise in a boolean
conditional (behavior-preserving; clears the 2 MAJOR reliability bugs)
- sqliteQuotaStore.poolUsage: drop the unreachable dimMap scan loops (dimMap
was never populated) — the lightweight snapshot already returns no
dimensions; poolUsageWithDimensions() is the plan-aware path
- BudgetTab: presentation role + keyboard handler on the checkbox wrapper
Integrated into release/v3.8.12. CodeQL hardening on the Chipotle executor: Math.random → crypto.randomInt/randomUUID, and a strict URL hostname check in the test. Fixed the node:crypto import (crypto.randomInt is not on the Web Crypto global → would crash at WS-connect) and added a regression guard exercising both helpers.
agy's gemini-3.1-pro-high/-low had no alias, so resolveAntigravityModelId sent the
speculative -high/-low suffix verbatim to upstream, which rejects it (400) for
gemini-3.x. Worse, the non-stream executor branch fed the 4xx response into the SSE
collector, returning a synthetic empty {object:chat.completion} envelope that masked
the error. Alias both to gemini-3.1-pro, and surface real upstream errors via
buildErrorBody for non-ok non-stream responses. + unit tests.
Residual of #3136: a local connection with an empty baseUrl still resolved
to this.config.baseUrl (OpenAI). Fall back to the provider's localDefault
(127.0.0.1:8080/v1) before the OpenAI default for the local-provider group.
Co-authored-by: tjengbudi <tjengbudi@users.noreply.github.com>
* chore(release): open v3.8.10 development cycle
Bump 3.8.9 → 3.8.10 across package.json, lockfile, electron, open-sse, and
docs/reference/openapi.yaml; add the [3.8.10] CHANGELOG section (root + 41 i18n
mirrors) as the integration target for the cycle. Entries land here as work
merges into release/v3.8.10; finalized by the release flow.
* fix(providers): resolve web provider alias collisions
Assign unique aliases to HuggingChat, Kimi Web, and Qwen Web so they no longer shadow primary providers or trigger startup warnings.
Add a unit test to enforce provider alias uniqueness and prevent future collisions. Also expand local ignore and VS Code exclude rules for agent, build, and worktree artifacts.
* fix(responses): normalize image_url parts across input paths (#3150)
Normalize image_url parts across all Responses input paths. Integrated into release/v3.8.10.
* fix(api-manager): preserve API key expiration local time (#3146)
Preserve API key expiration local time + clear button. Integrated into release/v3.8.10.
* Strip previous_response_id for stateless Responses upstreams (#3143)
Strip previous_response_id for stateless Responses upstreams (auto/strip/preserve). Integrated into release/v3.8.10.
* fix(opencode-plugin): map thinking cap to interleaved in model+combo (#3138)
Map caps.thinking to ModelV2.capabilities.interleaved for opencode-plugin. Integrated into release/v3.8.10.
* fix(providers): use synced models as fallback for all providers (#3148)
Use synced models as authoritative local catalog for all providers (+regression test). Integrated into release/v3.8.10.
* fix(qoder): bifurcate validation by token type — PAT→Cosy, regular API key→dashscope (#3149)
Bifurcate Qoder validation by token type (PAT→Cosy, regular→dashscope) +regression test. Integrated into release/v3.8.10.
* fix(antigravity): dynamic model resolution via MITM alias table (#3144)
Dynamic antigravity MITM model resolution in the executor (+bug fix +regression test; DB import dropped from client-reachable config). Integrated into release/v3.8.10.
* Feature/batch allow big (#3128)
Podman deployment options + larger upload body-size limits (+CONTAINER_HOST docs). Integrated into release/v3.8.10.
* fix(fireworks): preserve fully-qualified router/model IDs (#3133) (#3160)
Fireworks router IDs (accounts/fireworks/routers/...) were double-prefixed
with accounts/fireworks/models/ → upstream 404. Add optional
acceptedModelIdPrefixes to the registry entry and skip the prepend when the
model already starts with an accepted prefix.
Co-authored-by: KooshaPari <KooshaPari@users.noreply.github.com>
* fix(llama-cpp): route to configured local baseUrl instead of OpenAI (#3136) (#3161)
llama-cpp was missing from the local-provider group in buildUrl(), so it
fell through to the OpenAI baseUrl and returned an OpenAI 401. Add the
case to resolve the connection's providerSpecificData.baseUrl.
Co-authored-by: tjengbudi <tjengbudi@users.noreply.github.com>
* fix(t3-chat-web): parse cookies + convexSessionId from stored credential (#3007) (#3162)
The executor read credentials.cookies/convexSessionId, but the pipeline
only stores the pasted string under apiKey → t3.chat always 400'd. Parse
both values from apiKey (fallback accessToken), mirroring validation.ts.
Co-authored-by: minhtran162 <minhtran162@users.noreply.github.com>
* fix(minimax): stop capping MiniMax-M3 / M2.7 max_tokens at 8192 (#3141) (#3163)
MiniMax-M3 had no MODEL_SPECS entry and capitalized MiniMax-M2.7 missed
its lowercase spec (case-sensitive lookup) → both fell to the 8192 default
cap. Add the M3 spec (512K output), alias the capitalized ids, and make
getModelSpec lookups case-insensitive.
Co-authored-by: totaltube <totaltube@users.noreply.github.com>
* fix(github-copilot): discover model catalog live from api.githubcopilot.com (#3120, #3121) (#3164)
The github (Copilot) provider had a static hardcoded catalog with no
discovery source, so Import Models never refreshed (#3120) and advertised
non-entitled models that 400 on use (#3121). Add a live /models fetch with
fallback to the static list.
Co-authored-by: gabrielmoreira <gabrielmoreira@users.noreply.github.com>
* fix(combo): invalidate nested-combo cache on edits + log DATA_DIR (#3147) (#3165)
Editing a combo did not invalidate the 10s nested-combo expansion caches
(chat.ts getCombosCachedForChat + chatCore.ts getCombosCached; the exported
clearCombosCache was dead code), so a removed nested target/model could be
served as a phantom for up to 10s. Wire a shared monotonic combos-cache
version in readCache (bumped by invalidateDbCache("combos") on every combo
write); both cache layers treat a version mismatch as a miss.
Also log the resolved DATA_DIR/SQLITE_FILE absolute path at DB init so the
reporter's 'persists across restart + volume wipe' symptom (a multi-replica
Docker volume/DATA_DIR mismatch, not a routing bug) is diagnosable from logs.
Includes consolidated CHANGELOG entries for #3133/#3136/#3007/#3141/#3120/#3121.
Co-authored-by: ViFigueiredo <ViFigueiredo@users.noreply.github.com>
* fix(web-tools): parse bare JSON tool calls (#3157)
Parse bare JSON tool calls for deepseek-web (#2820) + fuzzy tool-name matching. Integrated into release/v3.8.10.
* fix(misc): minor fixes across reasoning cache, account fallback, binary manager (#3177)
Misc: ProviderProfile export, DeepSeek reasoning regex, binary guard. Integrated into release/v3.8.10.
* fix(kiro): minor OAuth social exchange tweaks (#3176)
Kiro social OAuth: optional targetProvider passthrough. Integrated into release/v3.8.10.
* deps: bump hono from 4.12.18 to 4.12.23 (#3179)
Bump hono to 4.12.23. Integrated into release/v3.8.10.
* fix(providerRegistry): update kilocode format and executor (#3166)
kilocode: openai format + default executor (matches kilo-gateway) + registry test. Integrated into release/v3.8.10.
* feat(metrics): cross-request TTFT and gap latency after tool calls (#3173)
Cross-request TTFT + gap-after-tool latency metrics (+test). Integrated into release/v3.8.10.
* feat(dashboard): provider stats API endpoint and dashboard page (#3175)
Provider stats dashboard + API (SQL moved to db module per Hard Rule #5, +test). Integrated into release/v3.8.10.
* fix(usage): sequential+spaced OAuth quota sync, reactive force-refresh, actionable 401 (#3156)
Sequential+spaced OAuth quota sync, reactive force-refresh on 401, actionable 401 in UI. Integrated into release/v3.8.10.
* fix(healthcheck): per-provider proactive-refresh skip list (rescue short-TTL OAuth) (#3159)
Per-provider proactive-refresh skip list (OMNIROUTE_HEALTHCHECK_SKIP_PROVIDERS) to rescue short-TTL OAuth. Integrated into release/v3.8.10.
* feat(quota): show OAuth token expiry on provider cards (small, blue, informative) (#3178)
Show OAuth token expiry on provider cards (small, blue, informative). Integrated into release/v3.8.10.
* fix(providers): empty refresh must not resurface just-cleared synced models (#3181)
Empty refresh must not resurface just-cleared synced models (fixes the release-blocking provider-models-route test). Integrated into release/v3.8.10.
* chore(release): v3.8.10 — 2026-06-04 (finalize CHANGELOG)
---------
Co-authored-by: Wilson <pedbookmed@gmail.com>
Co-authored-by: Xiangzhe <32761048+xz-dev@users.noreply.github.com>
Co-authored-by: Jan Leon <Jan.gaschler@gmail.com>
Co-authored-by: M.M <mr.maatoug@gmail.com>
Co-authored-by: Hernan Javier Ardila Sanchez <hjasgr@gmail.com>
Co-authored-by: Markus Hartung <mail@hartmark.se>
Co-authored-by: KooshaPari <KooshaPari@users.noreply.github.com>
Co-authored-by: tjengbudi <tjengbudi@users.noreply.github.com>
Co-authored-by: minhtran162 <minhtran162@users.noreply.github.com>
Co-authored-by: totaltube <totaltube@users.noreply.github.com>
Co-authored-by: gabrielmoreira <gabrielmoreira@users.noreply.github.com>
Co-authored-by: ViFigueiredo <ViFigueiredo@users.noreply.github.com>
Co-authored-by: PizzaV <103120356+pizzav-xyz@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Nicolas Lorin <androw95220@gmail.com>
* fix(sse): bound Antigravity short-retry 429 loop per endpoint
A persistent 429 on the short-retry branch (retryAfterMs ≤ 60s) looped
forever on the same endpoint because the branch did `urlIndex--; continue`
without checking the shared retry counter. Production log showed 77
consecutive 429s on one daily endpoint/account with zero fallback.
Gate the short-retry branch on `retryAttemptsByUrl[urlIndex] < MAX_AUTO_RETRIES`
(mirroring the already-bounded sibling), so a persistent 429 retries at most
3× per endpoint across all 3 base URLs then returns the 429 to the account-
fallback layer.
Regression test: 'bounds a persistent short-retry 429' in
tests/unit/executor-antigravity.test.ts — asserts 12 total attempts
(3 endpoints × 4) and a returned 429 with zero hang.
* fix(sse): lock Antigravity quota-exhausted account for full reset window
After the retry-loop bound, OmniRoute fell over to the next account but
re-selected the exhausted one first on every subsequent request (~60s wasted
per request). Root cause: the 429 body 'Individual quota reached. Contact
your administrator to enable overages. Resets in 164h27m24s.' was not
recognized as quota exhaustion, so the model was locked for only ~5s instead
of the real 6.8-day reset window.
Two detector fixes (mirrors Antigravity-Manager rate_limit.rs set_lockout_until):
1. classify429.ts — add QUOTA_PATTERNS: /individual quota reached/i,
/quota reached/i, /enable overages/i so looksLikeQuotaExhausted() fires.
2. accountFallback.ts — same patterns in classifyErrorText(); extend
parseRetryFromErrorText() to parse 'Resets? in XhYmZs' (reusing the
existing computeDurationMs helper) so the exact reset duration reaches
recordModelLockoutFailure as exactCooldownMs (uncapped, per user choice).
The lockout machinery already stores until = now + cooldownMs with no clamp,
bypasses getScaledCooldown when exactCooldownMs > 0, and keeps the longer of
existing/new, so the full 164h window flows through intact.
New patterns stay specific — plain 'too many requests'/'rate limit exceeded'
messages still classify as rate_limit.
Verified end-to-end against the real message:
- classify429 → quota_exhausted
- parseRetryFromErrorText → 592044000 ms (164h27m24s exactly)
- checkFallbackError → usedUpstreamRetryHint: true, cooldownMs: 592044000
* refactor(account-fallback): simplify error parsing and add cooldown safety
- Implement a 30-day cap on parsed retry durations to prevent indefinite account lockouts.
- Replace manual string matching with `looksLikeQuotaExhausted` for more robust quota detection.
- Streamline regex logic in `parseRetryFromErrorText` for better readability.
- Add unit tests for extreme cooldown values and free-tier exhaustion scenarios.
* fix(429): drop over-broad /quota reached/ pattern, keep specific matches
The bare /quota reached/ would also flag transient per-minute limits like
'request quota reached, retry in 60s' as quota_exhausted (multi-hour lock).
The Antigravity message is still caught by /individual quota reached/. Added a
regression assertion proving the transient case stays a rate_limit.
---------
Co-authored-by: diegosouzapw <diegosouza.pw@gmail.com>
* fix(sse): per-model 403 on passthrough providers locks the model, not the connection (#3027)
A per-model subscription 403 from a passthrough / per-model-quota provider
(e.g. ollama-cloud "this model requires a subscription, upgrade for access" on
deepseek-v4-pro) cooled down the ENTIRE connection instead of locking out only
the paid model, knocking out the free models on the same key and escalating an
exponential connection-wide backoff on repeats.
markAccountUnavailable's per-model lockout gate only covered 404/429/>=500, and
the only 403 -> model-lockout special case was hard-coded to grok-web. Generalize
it: a model-scoped 403 on an isPerModelQuotaProvider becomes a model lockout
(connection stays active). Terminal whole-key 403s (permanent/banned, account
deactivated, credits exhausted, project-route) keep their connection-level path.
Test-first: reproduction (paid model locked, connection active, free model still
eligible), regression guard (deactivated key still terminal -> banned, not
downgraded), and backoff guard (repeated 403s do not escalate connection backoff).
* feat(sse): persistent session + rolling-window memory for deepseek-web (#2942)
The DeepSeek web API takes only a single `prompt` string (no messages array) and
the executor created a fresh chat session per request, deleting it afterward — so
agentic multi-turn clients got per-turn amnesia.
Add two opt-in, per-connection settings (providerSpecificData), both defaulting to
the legacy behavior so plain-chat users are unaffected:
- historyWindow (number, default 0): when > 0, messagesToPrompt stitches the last N
non-system turns into a role-tagged transcript ("User:"/"Assistant:") so context
carries across turns within the single prompt string.
- persistSession (bool, default false): reuse one upstream chat session per userToken
(cached in the existing sessionCache) instead of creating/deleting one per request.
A reused session that fails is treated as stale (deleted in the DeepSeek UI): the
cache entry is dropped, a fresh session is created, and the completion is retried
once. Error paths invalidate the cached session so the next turn self-heals.
Test-first: pure prompt-builder window semantics (legacy/window/cap/empty) and
execute()-level session behavior via mocked fetch (fresh-per-request default, reuse,
stale-session fresh retry, history threaded into the prompt). Full existing
deepseek-web suite (35 tests) still green.
Note: chat.deepseek.com is an unofficial reverse-engineered surface; the live
round-trip can't be exercised in CI (no userToken/session). Treated as best-effort
per the issue; the prompt/session logic is unit-tested in isolation.
* feat(sse): tool-call translation for deepseek-web (#2820)
deepseek-web previously hard-failed any request carrying tools[] with a 400 (#2848),
so agentic clients could not use it for function calling at all. Add a bidirectional
translation layer (new open-sse/translator/webTools.ts, reusable by the other
web-cookie executors):
- Request: serializeToolsToPrompt() turns the OpenAI tools[] into a <tool>{...}</tool>
prompt contract, injected as a leading system message.
- Response: parseToolCallsFromText() extracts the model's <tool> blocks into OpenAI
tool_calls (arguments as a JSON string), strips them from content, and the executor
emits finish_reason "tool_calls" — for both non-stream and stream clients (the reply
is buffered since a tool block must be parsed whole).
A plain reply (no <tool> block) still streams normally with finish_reason "stop".
The superseded #2848 400-contract test is updated to assert the new translation.
Test-first: pure serializer/parser units + execute() round-trip via mocked fetch
(no-400, prompt serialization, non-stream tool_calls, stream tool_calls, plain reply).
Note: chat.deepseek.com is an unofficial reverse-engineered surface and the exact
<tool> emission depends on the model following the injected contract; the live
round-trip can't be exercised in CI. Best-effort per the issue; translation logic is
unit-tested in isolation.
* fix(sse): drop duplicate per-model 403 block — already in release via #3096
The release branch already scopes per-model 403 to model lockout (PR #3096,
commit 7042d562c) with the canonical !terminalStatus guard + exactCooldownMs
upstream hint. This PR's separate isTerminalOrRoute403 block shadowed it and
omitted the retry hint. Keep only the deepseek-web (#2942) + webTools (#2820)
changes here; the #3027 regression test is retained as coverage for #3096.
* feat(cursor): vision (image_url) input + tool-commit/output-constraint enhancements
Add image/vision input to the Cursor provider's agent.v1 endpoint, plus the
supporting prompt-engineering and resilience work developed alongside it.
Vision input
- Decode OpenAI `image_url` parts (base64 `data:` URIs and remote `http(s)` URLs)
and inline them as `SelectedContext.selected_images[]` — field numbers pinned
from the cursor-agent agent.v1 protobuf descriptor (SelectedImage.data oneof,
uuid, optional Dimension, mime_type). Cross-checked against composer-api's shape.
- New `resolveCursorImages` helper: SSRF-guarded remote fetches via the repo's
canonical `parseAndValidatePublicUrl` (always public-only for client URLs),
<=1 MiB per image (pre-decode + streaming cap), `image/*` enforced, max 12
images, sanitized `CursorImageError` (no stack/path leakage).
- `openai-to-cursor` translator now preserves `image_url` parts instead of
dropping them; executor `buildRequest` resolves images and attaches them to
the user turn. The no-image path is byte-identical to before (test-asserted).
Supporting cursor enhancements
- Tool-commit directive (raises composer-2.5 tool-call rate ~53% -> ~88%),
`tool_choice` none/required/specific handling, and output constraints
(`response_format` / `max_tokens` / `stop` surfaced as prompt instructions).
- `cursorSessionManager`: clear pending tool-call mappings on session close.
- `cursorVersionDetector`: export `FALLBACK_VERSION` as a single source of truth.
Tests & docs
- New unit suite for the image encoder + resolver (field layout, byte-identical
no-image path, SSRF / oversize / bad-base64 / too-many rejections, sanitized
error body), translator image-preservation tests, and live e2e tests
(base64 + remote URL, gated on `CURSOR_E2E_TOKEN`).
- Documented `CURSOR_TOOL_DIRECTIVE` and `CURSOR_IMAGE_FETCH_TIMEOUT_MS` in
`.env.example` and `docs/reference/ENVIRONMENT.md`.
* fix(cursor): address review — redirect SSRF, large-payload guard, stream OOM, case/NaN nits
Resolves the gemini-code-assist review on #3104:
- SSRF via redirect (critical): fetchImageBytes now uses redirect:"manual" and
re-validates every hop through parseAndValidatePublicUrl, so a public URL can't
30x-redirect to a private/link-local address. Bounded to 3 redirects.
- Large data URL (high): reject on raw payload length before the whitespace-strip
regex, so an oversized data URL can't burn CPU.
- Stream read (high): readCapped consumes the body as an async iterable (Node
Readable + Web Streams) or via getReader, capping mid-read; uncapped
arrayBuffer() is only a last resort.
- data: scheme (medium): match case-insensitively (RFC 2397) while preserving the
original payload.
- NaN timeouts (medium): CURSOR_IMAGE_FETCH_TIMEOUT_MS and CURSOR_STREAM_TIMEOUT_MS
fall back to defaults when the env value isn't a positive integer.
Adds tests: redirect-to-private blocked, redirect-to-public followed, too-many-
redirects rejected, uppercase DATA: accepted.
* fix(cursor): defend image fetch against DNS-rebinding SSRF
Address the @codex review on #3104: parseAndValidatePublicUrl only checks the
hostname string, so a public-looking host that (re)resolves to a private /
link-local / metadata IP would still be fetched. Each hop now resolves the host
via dns.lookup({all:true}) and rejects if ANY answer is private (isPrivateHost),
before connecting. IP literals are skipped (already validated by the URL guard).
This narrows but doesn't fully close the TOCTOU window vs fetch's own
resolution; a connection-time IP filter on the shared outbound guard would
close it for every caller. Adds unit tests for the IP gate and a mocked
DNS-rebinding case (public host -> 127.0.0.1, fetch never reached).
grok-web only sent the 'sso' cookie; Grok's anti-bot now rejects (403 code 7) requests missing the paired 'sso-rw' write cookie. Adds buildGrokCookieHeader() which emits sso plus sso-rw when the pasted blob carries it (never a phantom sso-rw from a bare value), used by both the executor and the connection validator. Updates the grok-web authHint. Tests: 6 new buildGrokCookieHeader cases; grok-web 62/62 unchanged.
Add a global OMNIROUTE_CODEX_WS_ENABLED kill-switch for Codex WebSocket transport, defaulting to enabled if feature flag lookup fails.
Track Xiaomi MiMo monthly token usage from OmniRoute usage history and expose it through the usage fetcher provider list.
#3054 ("remove 9 dead/unreachable free providers") removed the petals/nanobanana
configs, registry entries and validators but left dangling references that broke
the build and the unit suite on release/v3.8.8:
- open-sse/executors/petals.ts imported the deleted ../config/petals.ts
(webpack "Module not found" → `next build` failed). Removed the executor, its
registration + re-export in executors/index.ts, and the leftover
`providerId === "petals"` branch in providerAllowsOptionalApiKey.
- Removed tests for the now-deleted providers: executor-petals.test.ts and
poolside-provider.test.ts (REGISTRY.poolside was removed), and the petals /
nanobanana validator assertions in provider-validation-specialty.test.ts,
plus the stale petals catalog assertions in providers-page-utils.test.ts,
proxy-connection-test.test.ts and providers-route-managed-catalog.test.ts.
The image/video/embed registries for nanobanana/replicate/nomic are real and
untouched — only the dead chat/api-key surfaces were removed. 146/146 affected
tests pass; typecheck / build clean.
* fix(pollinations): wire session pool + add idle pruning
PollinationsExecutor.getPool() returned null because poolConfig was
never set. Now sets DEFAULT_POOL_CONFIG in constructor so anonymous
requests get fingerprint rotation and 429 cooldown management.
Also:
- Use acquireBlocking() instead of acquire() to wait for available session
- Add startAutoPrune() for periodic idle session cleanup (5min idle timeout)
- Improve execute() error handling with proper finally block
* fix(duckduckgo-web): wire session pool for fingerprint rotation
DuckDuckGoWebExecutor had poolConfig set but never called getPool()
in execute(). Added session acquisition via acquireBlocking() and
merges fingerprint headers into fetch calls for rate limit evasion.
* fix: address PR #3049 review comments
- duckduckgo-web: fix session leak (add finally release), fix race
condition (report status before release), remove duplicate sessionHeaders
- pollinations: fix race condition (move release to finally, report before)
* fix: address all PR #3049 review comments
- duckduckgo-web: add pool reporting on 429/500 early returns (was missing)
- duckduckgo-web: add sessionHeaders to retry fetch on 401/403
- sessionPool: add .unref() to setInterval to prevent keeping process alive
---------
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>
The execute() and testConnection() methods called
normalizeClaudeSessionCookie() which does NOT inject cf_clearance.
Cloudflare pins cf_clearance to TLS fingerprint — without it, requests
get challenged with 403 even with valid session cookies.
Changed both call sites to use normalizeClaudeSessionCookieWithAutoRefresh()
which auto-injects cf_clearance via Turnstile solver when missing.
Fixes: [403]: Claude Web API error:
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>
- Bump CLAUDE_CLI_VERSION / CLAUDE_CODE_VERSION from 2.1.146 to 2.1.158
(matches claude-cli 2.1.158 captures)
- Add redact-thinking-2026-02-12 to the always-sent beta tier
(present in Haiku, Sonnet, and Opus CC 2.1.158 captures)
- Add mid-conversation-system-2026-04-07 for Opus full-agent only
(present in Opus capture, absent from Sonnet/Haiku)
- Remove afk-mode-2026-01-31 from heavy-agent beta flags
(not present in any real Claude Code 2.1.158 capture)
- Add mid-conversation-system-2026-04-07 to ANTHROPIC_BETA_BASE
- Update CC-compatible bridge version strings
- Fix STEALTH_GUIDE.md outdated version + Stainless version (0.81.0 → 0.94.0)
- Update claude-beta-flags-2454 tests to match real CC behavior
Pollinations retired the legacy text.pollinations.ai host, which now returns
404 'This is our legacy API' for all models. The registry primary baseUrl (and
the executor's hardcoded fallback) still pointed at it; gen.pollinations.ai/v1
is the current OpenAI-compatible gateway (it was already listed as the
secondary fallback). Make gen the sole/primary endpoint and align the executor
test.
Closes#2987
Free-plan Codex accounts (workspacePlanType === "free", from the OAuth id_token)
cannot run the server-side image_generation hosted tool, but the Codex CLI
injects it into every Responses request — causing an upstream 400. normalizeCodexTools
now drops image_generation when the connection is a free-plan account; paid plans
keep it. Mirrors CLIProxyAPI's isCodexFreePlanAuth guard.
Found during the #2980 (Codex OAuth compat) analysis against CLIProxyAPI.
- duckduckgo-web.ts: Add poolConfig (2-5 sessions, 1s cooldown)
- providerRegistry.ts: Add poolConfig field to RegistryEntry type, configure for llm7
- default.ts: Read poolConfig from registry in constructor
DuckDuckGo Web now uses pool for VQD token rotation.
LLM7.io now uses pool with 2s cooldown for its 1 req/s rate limit.
Move pool integration from Pollinations-specific code into BaseExecutor
so any executor can opt in via a protected poolConfig property.
Changes:
- base.ts: Add poolConfig, getPool(), buildPoolHeaders() to BaseExecutor
- pollinations.ts: Remove static pool, use base class pattern
- session-pool-modular.test.ts: 36 tests for provider-agnostic pool behavior
Closes#2953
The agy provider was registered in providerRegistry.ts with
executor: "antigravity" but the executor map in executors/index.ts
only had an "antigravity" entry. getExecutor("agy") fell through to
DefaultExecutor, which returned undefined for baseUrl (agy only has
baseUrls), causing fetch(undefined) → TypeError: Cannot read properties
of undefined (reading 'toString').
Closesdiegosouzapw/OmniRoute#2932
